Pedestrian Detection with YOLOv5 in Autonomous Driving Scenario

Abstract

Autonomous vehicle, with the attributes that ensuring driving safety and improving traffic efficiency, has been a research hotspot for a long time. In the modular developing pipeline of autonomous vehicles, pedestrian detection based on computer vision is a critical component of perception module. In this paper, we apply the newly proposed network structure YOLOv5 in pedestrian detection problem. After training in PASCAL VOC2012 dataset, the model realizes high detection accuracy and real-time efficiency. At the same time, the model owns competitive generalization ability which achieve high detection accuracy in KITTI dataset. With competitive detection accuracy and real-time efficiency, YOLOv5 have the potential to be deployed on autonomous vehicles.